1. List and briefly describe the activities that a router or switch
is likely to engage in to provide a certain QOS guarantee in a network.
(i)
Policing: verifying that
incoming traffic conforms to its agreed specification
(ii)
Admission control: checking to
see if there are enough resources to meet a request for QoS
(iii)
Classification: recognizing
those packets that need particular levels of QoS
(iv)
Queuing and scheduling: making
decisions about when packets are transmitted and which packets are dropped that
are consistent with the QoS guarantees
2. What is the primary difference between IntServ and DiffServ
approaches to allocating resources for QOS purposes ? Which approach scales
better for a large IP network and why ?
IntServ enables end-to-end QoS guarantees for a single flow
(microflow) from one sender to one or more receivers through a signalling
protocol like RSVP
DiffServ a coarse-grained model groups traffic of many different
flows into several small classes and provides QOS guarantees on a per-class
basis
DiffServ scales better. In a large IP network, the number of
microflows is very large and to keep QOS and state information about each one
provides a heavy burden on the routers in the network, making it difficult to
scale. Grouping flows into classes allows easier management of QOS information.
3. In what way does the deficit round robin (DRR) approach provide
for a more fairer QOS allocation of transmission resources on a router as compared
to the weighted round robin approach ? Briefly explain how it works.
It computes weights allocated to different flows in terms of total
bytes rather than number of packets. This is fairer QOS allocation because IP
packets vary in size
DRR allocates a number of bytes to each flow. When a flow is
selected, DRR transmits as many packets as possible without exceeding the
allotted number of bytes.
The remainder (i.e. the difference between the number of bytes that
was allocated and the size of the packets actually sent) becomes a deficit
which is added to the amount that will be sent in the next round
4. Explain briefly how RSVP works with reference to the PATH and
RESV messages.
(i)
PATH messages travel from a
sender to one or more receivers (explicitly designed to support multicast) and include TSpecs and classification
information provided by the sender.
(ii)
When a receiver gets a PATH
message, it can send a RESV message back toward the sender. The RESV message
identifies the session for which the reservation is to be made and indicates
the level of QoS required by this receiver.
(iii)
Messages are intercepted by
every router along the path, so that resource allocation can take place at all
the necessary hops.
(iv)
Each router must agree to
reserve the resources the request specifies. Reservation is unidirectional for
a single flow direction
5. Three clients are receiving shared multicast video traffic flow across
a MPLS network from a media server. The routes from these different client
converge on a router that the server is directly connected to. All 3 clients
make a QOS reservation requests for a delay latency of 30ms, 15 ms and 10 ms
respectively to this router.
(i)
How would RSVP handle these
requests in the router ?
(ii)
How does RSVP ensure that these
reservation requests are propagated correctly across the MPLS network to this
router at the transmitting end ?
i) These requests are merged into a single reservation request for
the common flow, to which the router will select the smallest latency of 10ms
ii) When a path message in RSVP is sent to a client, each router at
each hop inserts its own IP address as the message’s last hop. Each router can
look at the last hop field to learn where the flow came from. Should it later
receive a reservation request for this flow, this last hop information tells it
where to send the reservation request next.
6. Explain briefly what the term soft state means within the context
of RSVP. Consider an ongoing video conference involving 2 clients receiving
multicast traffic from a central server.
A new client wishes to join this multicast group. Outline the series of steps
that it will undergo to achieve this, and show how soft state helps in this
registration process.
Soft state means that PATH and RESV messages in RSVP must be sent
periodically to refresh a reservation. If they are not sent for some interval
(the time-out period), then the reservation is automatically torn down
- The client
sends an IGMP group membership report message to join this multicast
group.
- The local
router receives this message and to update the rest of the network, the router
builds an OSPF link state advertisement and floods it through the network.
Once all routers are updated,
multicast traffic begins to flow to the new client.
- Path
messages for the flow can reach it as well, and soft state assures that a periodic
path message from the sender will eventually reach the new client.
- With this
path message in hand, the client can identify the flow and place its own
reservations.
7. How does label distribution
in MPLS assist in the creation of a RSVP reserved flows ? Describe briefly the
steps involved in this process which lead to the creation of a label switched
path (LSP).
Labels are bound between flows that have RSVP reservations (through
the use of a FEC) and then distributed through the MPLS network.
When an LSR wants to send a RESV message for a new RSVP flow:
(i)
It allocates a label from its
pool of free labels
(ii)
Creates an entry in its LFIB
with the incoming label set to the allocated label
(iii)
Sends out the RESV message
containing this label to the upstream LSR.
(iv)
Upon receipt of a RESV that
contains this label, an LSR populates its LFIB with this label as the outgoing
label
(v)
It then allocates a new label
to use as the incoming label and inserts that in the RESV message before
sending it upstream.
(vi)
As RESV messages propagate
upstream, an LSP is established along the path.
8. What is meant by Per Hop Behaviour (PHB) in the context of the
DiffServ approach to QOS allocation and how is it indicated ? Briefly describe
the 3 main types of PHBs
DiffServ allows each node along the path to define the service that
a given class will receive and this can vary from node to node along a given
path, therefore DiffServ does not provide QOS end-to-end guarantees. Each of
the standard PHBs is indicated by a recommended DSCP value, and each router has
a table that maps the DSCP found in a packet to the PHB that will determine how
that packet is treated
Default - No special treatment, equivalent to best effort.
Expedited forwarding (EF) - Packets marked EF should be forwarded
with minimal delay and experience low loss.
Assured forwarding (AF) - Packets go into different priority queues
where they have different drop preferences if congestion occurs
9. Briefly describe the standard approach towards congestion
avoidance in TCP and state two disadvantages with this approach. Explain how Explicit Congestion Notification
(ECN) overcomes these disadvantages.
Congestion avoidance in TCP assumes that packet losses in the network
are an indication of congestion, and reduce their sending rates when they
experience packet loss.
Disadvantages:
(i)
For
real time applications, the lost packet will need to be transmitted and its
late arrival will cause degradation
(ii)
A
lost packet consumes resources up to the point it is lost, it would be
preferable not to send the packet at all if it is just going to be thrown away
Router sets “congestion experienced” (CE) bit in packet header when
it detects congestion, and then forward the packets rather than dropping it.
The router must have some form of queue management to actively monitor
congestion. When a packet with the CE bit arrives at its destination, the
receiver must send a signal back to the sender that will cause the sender to
reduce its sending rate.
10. A multimedia network that provides QOS guarantees uses a leaky
bucket policer in one of its routers to ensure that the incoming packet traffic
does not exceed the TSpec specification agreed upon during an initial session
of Integrated Services (IntServ). The following are the features of this
policer:
·
The token buffer can hold at
most three (3) tokens, and is
initially filled with two (2) tokens at
time slot t = 0.
·
New tokens arrive into the
bucket at a rate of two (2) tokens per
time slot. Packets arrive at the beginning of a time slot and enter the
packet queue, where they are processed and transferred to the output link in a First In First Out (FIFO) manner.
·
The size of the packet queue is
four (i.e. it can queue a maximum of 4 packets at any given time slot); any extra arriving
packets are dropped.
·
Packets that obtain available
tokens in a given time slot go together
on the same time slot in the output link.
Time slot
|
Incoming Packets
|
0
|
A B C D
|
1
|
E F
|
2
|
G
|
3
|
-
|
4
|
-
|
5
|
H I J K
|
6
|
L M N O
|
7
|
P Q
|
8
|
-
|
9
|
R S T
|
The table shows incoming packets from the network into the router
with the policer, from time slot t = 0 to time slot t = 9. Based on this
information, construct a new table with columns showing the packets in queue,
tokens in bucket and packets on output link from time slot t = 0 to t = 9.
Time slot
|
Packets in queue
|
Tokens in bucket
|
Packets at output
|
0
|
A B C D
|
2
|
A B
|
1
|
C D E F
|
2
|
C D
|
2
|
E F G
|
2
|
E F
|
3
|
G
|
2
|
G
|
4
|
-
|
3
|
-
|
5
|
H I J K
|
3
|
H I J
|
6
|
K L M N
|
2
|
K L
|
7
|
M N P Q
|
2
|
M N
|
8
|
P Q
|
2
|
P Q
|
9
|
R S T
|
2
|
R S
|
